kennss/SiliconScope

Most macOS system monitors tell you CPU and GPU load and call it done. This open-source project goes further for Apple Silicon specifically, exposing the Neural Engine, Media Engine, and memory bandwidth metrics that matter when you are running local models or video pipelines. The native SwiftUI implementation means it is not a cross-platform app with a macOS skin — it is built for the platform, and the sudoless constraint means no sudo prompts or security trade-offs to get hardware-level data. For developers running llama.cpp, Whisper, or any Core ML workload, seeing ANE utilization alongside memory bandwidth in one view is genuinely useful for understanding whether your model is actually hitting the accelerator or falling back to CPU. Reservation: this is a young repo with a modest track record, and the feature set is still narrow — do not expect Instruments-level depth. Treat it as a lightweight always-on companion, not a profiling replacement. -> Best for: ML researcher or indie hacker running local inference on Apple Silicon